I realize that FC4 is unexciting these days, so this may not generate much enthusiasm, but: how in the _world_ are you RH folks building updates for Fedora Core 4?
The story:
We need to make some local modifications to some of the files in the 'setup' package here at BU, and we were chasing ourselves in circles over making it not conflict with /var/log/lastlog also owned by util-linux.
Both of these packages have that file as a %ghost file, and last September, they were updated to have permissions of 0644. Specifically, there's
%ghost %attr(0644,root,root) %verify(not md5 size mtime) /var/log/lastlog
in both files.
Okay, fine, that should work. And the packages in the FC4 updates area indeed both install without conflict.
But then, when we went to install our package, bam:
file /var/log/lastlog from install of setup-2.5.44-1.1bu46.1 conflicts with file from package util-linux-2.12p-9.14
Eh? But the entries are the same. After much trying-things-to-no-avail, I tried just plain rebuilding the setup package with no changes.
file /var/log/lastlog from install of setup-2.5.44-1.1 conflicts with file from package util-linux-2.12p-9.14
Hey! It's not anything we're changing -- rebuilding the unmodified package doesn't work either.
Diagnosing this was a bit complicated by the fact that ghost files don't show up in rpm -ql, but eventually with a bit of python, I dumped the md5sums of /var/log/lastlog stored in each RPM. Ah-ha -- that's the problem. The package I build comes out with nothing. (Either 0 or "" -- not precisely sure which and it doesn't really matter...) The RPM built by RH in the FC4 update tree has "d41d8cd98f00b204e9800998ecf8427e" -- the md5sum of an empty file.
Well, that's weird, I thought. Maybe it's because we're using the updated version of RPM -- rpm-4.4.1-22 -- instead of the one that shipped with FC4 originally.
No, that's not it -- that version also results in 0 md5sums for ghost files. I can't find exactly where this is in the RPM changelog, but I had to go back to the FC2 version of rpm to build a package with the empty-file checksum as in the FC4 update.
So, what th' heck? Does your FC4 build system use rpmbuild from FC2, or what? :)
We're working around this by rebuilding util-linux too, which works, but the whole thing was a headache and I don't like how this could happen with other packages.....
Matthew Miller wrote:
I realize that FC4 is unexciting these days, so this may not generate much enthusiasm, but: how in the _world_ are you RH folks building updates for Fedora Core 4?
The story:
We need to make some local modifications to some of the files in the 'setup' package here at BU, and we were chasing ourselves in circles over making it not conflict with /var/log/lastlog also owned by util-linux.
Both of these packages have that file as a %ghost file, and last September, they were updated to have permissions of 0644. Specifically, there's
%ghost %attr(0644,root,root) %verify(not md5 size mtime) /var/log/lastlog
in both files.
Okay, fine, that should work. And the packages in the FC4 updates area indeed both install without conflict.
In my experience, that does *not* work. Sharing of files (whether their %ghost'd or not) *only* works if the original contents are *exactly* the same (timestamp, ownership, permissions, contents, md5sum, etc...).
-- Rex
On Fri, Mar 31, 2006 at 06:48:08AM -0600, Rex Dieter wrote:
%ghost %attr(0644,root,root) %verify(not md5 size mtime) /var/log/lastlog in both files. Okay, fine, that should work. And the packages in the FC4 updates area indeed both install without conflict.
In my experience, that does *not* work. Sharing of files (whether their %ghost'd or not) *only* works if the original contents are *exactly* the same (timestamp, ownership, permissions, contents, md5sum, etc...).
You got util-linux and setup installed on your system? It's working. :) Timestamps can be different, but yeah, everything else should be the same.
In the case of the problem described, the files _are_ the same -- they're zero length files. But, as far as I can tell, FC4's RPM stores the md5sum of ghosted files differently from whatever version of RPM Red Hat is using to actually build the FC4 packages they release.
In an RPM built on FC3 and above, the package contains _nothing_ for the md5sum. In packages built on FC2, and to my surprise, the FC4 packages in the updates repository, the ghosted files have their "correct" checksum -- that of an empty file.
That's fine as long as you don't try to rebuild any of those packages locally, but if you do, you'll run into this issue.
I assume the change was made so that ghosted files _don't_ have to start identically, which makes sense, but it causes weirdness when the official packages aren't built that way.
-
Matthew Miller -
Rex Dieter